A naked-eye 3D visualization method for gradient batch parallel multi-resolution 3D modeling of clothing patterns
By establishing a gradient parameter system for clothing patterns and parallel task allocation, combined with real-time naked-eye 3D display feedback, the efficiency and consistency issues of 3D modeling of clothing patterns in existing technologies have been solved, and efficient multi-resolution 3D modeling has been achieved.
Patent Information
- Application Number
- CN202611552690.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-09
- Publication Date
- 2026-11-17
AI Technical Summary
Existing 3D modeling methods for garment patterns struggle to achieve sufficient accuracy when dealing with differences in the complexity of structures in different regions. This results in insufficient detail in complex areas, excessive redundant meshes in simple areas, low efficiency in batch tasks, and long parameter adjustment feedback loops, all of which affect model generation efficiency and consistency.
Using naked-eye 3D visualization technology, by establishing a gradient parameter system for clothing patterns, the complexity of different structural areas is distinguished. Parallel task allocation and load adjustment are adopted, combined with real-time display feedback to adjust parameters, to realize the construction and splicing of multi-resolution meshes, forming a closed-loop modeling process.
It improves the generation efficiency and detail of multi-specification clothing 3D models, shortens the waiting time for batch tasks, enhances gradient consistency, and optimizes the allocation of modeling resources.
Smart Images

Figure CN123118926A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of 3D modeling technology, and in particular to a naked-eye 3D visualization method for gradient batch parallel multi-resolution 3D modeling of clothing patterns. Background Technology
[0002] The field of 3D modeling technology involves using computer graphics processing techniques to construct spatial models of objects, organize geometric data, and generate 3D visual representations. Among these, the gradient batch parallel multi-resolution 3D modeling method for clothing patterns refers to a technique based on a computer-aided modeling system to transform the spatial structure of clothing pattern data. This typically involves a processing flow including 3D coordinate data acquisition, surface mesh construction, texture data mapping, and model file storage. The conversion from 2D pattern data to 3D model data is completed using a graphics processor or computer graphics software, and a model database is used to store 3D structural information at different resolutions.
[0003] Existing 3D modeling of garment patterns typically involves meshing the overall structure with uniform precision. However, the differences in structural complexity across different regions are difficult to fully reflect in the allocation of modeling resources. Complex detail areas are prone to insufficient representation, while simple outline areas are prone to excessive meshing. This results in a high overall computational burden. When batch pattern tasks are executed sequentially, processing delays are likely to occur. Furthermore, repeated parameter configuration and model reconstruction are required between different specifications. The feedback loop between preview results and parameter adjustments is long, which affects the efficiency of generating multi-specification garment models and the consistency of local representation. Summary of the Invention
[0004] The purpose of this invention is to provide a naked-eye 3D visualization method for batch parallel multi-resolution 3D modeling of clothing patterns. By establishing a gradient parameter system for clothing patterns, the structural data of different clothing categories, sizes, and styles are organized into gradient modeling tasks. Combined with naked-eye 3D display, the complexity of different structural regions is differentiated, and different regions are constructed with 3D meshes according to the corresponding modeling resolution. At the same time, a parallel task allocation and load adjustment mechanism is adopted to complete batch modeling. The gradient parameters of the pattern and the modeling accuracy of the region are adjusted based on the real-time display results of naked-eye 3D. Thus, a 3D modeling process with interconnected resolution configuration, parallel modeling, and display correction for different gradient patterns and regions with different structural complexities is formed.
[0005] To achieve the above objectives, this invention provides a naked-eye 3D visualization method for gradient batch parallel multi-resolution 3D modeling of clothing patterns, comprising the following steps: The process involves acquiring the pattern structure data of the garment to be modeled, dividing it into gradient levels according to garment category, size specifications, and pattern style, establishing a pattern gradient parameter system, and constructing pattern gradient modeling tasks based on the pattern gradient parameter system and pattern structure data. The pattern structure data corresponding to each pattern gradient modeling task is then displayed in naked-eye 3D to determine the structural complexity of the overall outline area, detail texture area, sewing structure area, and fabric fold area. Based on the structural complexity, high, medium, or low modeling resolutions are matched to each area to determine the area modeling accuracy configuration. Based on the area modeling accuracy configuration, each pattern gradient modeling task is allocated to multiple parallel threads. Three-dimensional meshes for different areas are constructed and stitched together according to the corresponding modeling resolution to generate gradient pattern 3D models. Each gradient pattern 3D model is then displayed in real-time using naked-eye 3D. The pattern gradient parameter system and area modeling accuracy configuration are adjusted based on the display results, and the corresponding pattern gradient modeling tasks are re-executed.
[0006] Furthermore, when establishing the pattern gradient parameter system and pattern gradient modeling task, the contour control points, seam boundaries, and size reference values in the garment pattern are obtained. Size differences between different size specifications are determined using the reference specifications as a reference, and the shape adjustment amount of the contour parts is determined according to the pattern style. A parameter mapping relationship is established between the gradient level and the size difference and shape adjustment amount. The size difference and shape adjustment amount are written into the corresponding contour control points and seam boundaries to form the pattern structure data corresponding to each gradient level. The garment category, gradient level, and pattern structure data are then linked to form a task identifier.
[0007] The size difference is determined by comparing the target size specification with the baseline size specification in terms of bust, waist, hip, garment length, and sleeve length, and allocating the corresponding differences to the contour control points according to the grading direction of the garment category. The shape adjustment amount is determined by the applicable adjustment items among the looseness allowance, dart allowance, and silhouette offset corresponding to the pattern style, ensuring that the cumulative adjustment amount of the contour area is within the displacement range that maintains continuous splicing of adjacent seam boundaries. The looseness allowance is determined by the difference between the finished garment size and the net body size of the same part; the dart allowance is determined by analyzing the dart endpoints and accumulating the distance between the endpoints on both sides of the dart along the corresponding seam boundary; the silhouette offset is determined by calculating the normal displacement of the pattern style's corresponding contour control point relative to the baseline pattern contour at the same longitudinal reference position and recording the displacement direction.
[0008] Furthermore, when generating the region modeling accuracy configuration, the contour boundaries, texture boundaries, sewing lines, and fold ridges are converted into spatial line elements and mapped to the naked-eye 3D display coordinate system. Based on the structural attributes of these spatial line elements, they are categorized into the overall contour region, detail texture region, sewing structure region, and fabric fold region, respectively. The line element density, direction change count, overlap count, and depth change count for each region are recorded. Complexity indices are selected based on the actual structural attributes of each region. The complexity indices in the same pattern gradient modeling task are converted to a unified numerical range according to their maximum values and then merged to determine the structural complexity. Finally, the regions are sorted and their modeling resolutions are matched according to their structural complexity from high to low.
[0009] For the overall contour area, the complexity index includes the density of control points and the density of directional changes per unit contour length; for the detailed texture area, the sewing structure area, and the fabric fold area, the complexity index is calculated based on the texture boundary length and the number of local depth abrupt change positions, the sewing line path density and the number of path intersection positions, the fold ridge and valley line density and the number of local depth extreme position positions, and converted into a feature density per unit area according to the corresponding area area, so as to form the corresponding complexity index.
[0010] To further limit the modeling resolution, the spacing between adjacent contour control points, the minimum spacing of texture boundaries, the spacing between adjacent sewing lines, and the spacing between adjacent fold ridges and valleys are obtained from the pattern structure data. The minimum spacing is selected from the actual existing spacing data, and the grid target side length that is no more than half of the minimum spacing is determined as the high-resolution reference side length. The medium-resolution target side length and the low-resolution target side length are determined by two times and four times the high-resolution reference side length, respectively. The corresponding target side lengths are written into the modeling accuracy parameters of the corresponding area.
[0011] Furthermore, during parallel modeling, the expected number of mesh cells is determined based on the region area and the target mesh side length. Then, according to the number of mesh cells to be processed by each parallel thread, the region to be processed is allocated to the thread task queue with the lowest load. Each parallel thread reads the boundary points and internal structure lines of the corresponding region, determines the target mesh side length according to the corresponding modeling resolution, and performs triangular meshing to generate a 3D region mesh. For spatially adjacent 3D region meshes, their common boundary vertex sequence is obtained, vertices at the same position are aligned, and transition mesh cells are added. The 3D region meshes with aligned boundaries are merged to form a gradient-patterned 3D model.
[0012] For parallel thread load adjustment, the thread load value is determined based on the sum of the expected number of grid cells in the allocated regions of each thread's task queue, and the regions to be allocated are written to the thread task queue with the lowest thread load value. The corresponding thread load value is updated after the construction of a 3D mesh for each region is completed. When the deviation of any thread load value from the average load value of all parallel threads exceeds a preset load deviation threshold, the region with the largest expected number of grid cells that has not yet been processed in the highest-load queue is moved to the lowest-load queue, and the task thread mapping relationship is updated.
[0013] Furthermore, when displaying the gradient pattern 3D model in real-time using naked-eye 3D, real-time view frames of the model from multiple display viewpoints are acquired. The pattern structure data is projected according to the corresponding display viewpoints to form contour baselines and stitching baselines. The positional deviations of the model contour and stitching boundaries relative to the contour baselines and stitching baselines are calculated, and the local surface dispersion is determined based on the display depth jumps of adjacent mesh surfaces. When the positional deviation exceeds a preset deviation threshold, the pattern gradient parameter system of the corresponding gradient level is adjusted; when the local surface dispersion exceeds a preset dispersion threshold, the modeling resolution of the corresponding area is increased and the area modeling accuracy configuration is updated, and the corresponding pattern gradient modeling task is re-executed.
[0014] After each re-execution of the pattern gradient modeling task, the contour position deviation, stitch boundary position deviation, and local surface dispersion of the current round are compared with those of the previous round. When all contour position deviations and stitch boundary position deviations are not greater than the preset deviation threshold, and all local surface dispersions are not greater than the preset dispersion threshold, the re-execution of the pattern gradient modeling task is stopped.
[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: Modeling tasks for different categories, sizes, and styles of clothing are organized around a gradient parameter system for garment patterns. Based on the results of naked-eye stereoscopic display, the complexity of areas such as overall outline details, textures, sewing structures, and fabric wrinkles is identified. This allows the modeling accuracy to be allocated according to the expression needs of each area, preserving structural details in complex areas and compressing redundant meshes in simple areas. At the same time, different gradient tasks are assigned to parallel threads to complete multi-resolution mesh construction and stitching, shortening the waiting time for batch tasks. Combined with real-time stereoscopic display feedback, the pattern gradient parameters and area accuracy configuration are adjusted in a coordinated manner, forming a closed loop between model reconstruction and visual verification. This improves the generation efficiency, detail representation, and gradient consistency of 3D models of multi-size clothing. Attached Figure Description
[0016] Figure 1 A schematic diagram of the overall execution chain of a method for naked-eye 3D visualization of clothing patterns using gradient batch parallel multi-resolution 3D modeling. Figure 2A schematic diagram illustrating the data mapping relationship between the pattern gradient parameter system and the pattern gradient modeling task; Figure 3 A schematic diagram illustrating the relationship between the complexity of different structural regions and the modeling accuracy of multi-resolution regions; Figure 4 A schematic diagram illustrating the relationship between parallel thread task mapping and thread load adjustment; Figure 5 This is a schematic diagram illustrating the relationship between viewpoint deviation feedback and re-execution state in naked-eye 3D multi-display. Detailed Implementation
[0017] This invention is used for gradient 3D modeling of garment patterns. The modeling input is the pattern structure data of the garment to be modeled. The pattern structure data includes at least contour control points, seam boundaries, and dimensional reference values; when entering the region precision processing, contour boundaries, texture boundaries, seam lines, and fold ridge lines are also used. All types of data are used within the same pattern gradient modeling task, and data from different gradient levels are distinguished by task identifiers.
[0018] The pattern gradient parameter system records the correspondence between gradient levels, size differences, and shape adjustments. Size differences are formed by the differences in bust, waist, hip, garment length, and sleeve length between the target size specification and the baseline specification, and are allocated to the contour control points according to the grading direction of the corresponding garment category. Shape adjustments are formed by the actual applicable adjustments among the looseness allowance, dart allowance, and silhouette offset corresponding to the pattern style. When shape adjustments are written into the contour control points and seam boundaries, their cumulative displacement does not exceed the displacement range within which adjacent seam boundaries can maintain continuous splicing.
[0019] The region modeling precision configuration records the modeling resolution and target mesh edge length for the overall contour region, detailed texture region, sewing structure region, and fabric fold region. The expected number of mesh cells is determined by the region area and target mesh edge length and is used for thread task allocation. The display deviation record saves the contour position deviation, seam boundary position deviation, and local surface dispersion, used to determine whether to re-execute the current pattern gradient modeling task. Example
[0020] Please see Figure 1 This embodiment provides a naked-eye 3D visualization method for batch parallel multi-resolution 3D modeling of clothing patterns using gradients. The method is executed according to steps S1 to S4. S1 outputs the pattern gradient modeling task, S2 outputs the region modeling accuracy configuration, S3 outputs the gradient pattern 3D model, and S4 reads the gradient pattern 3D model and corresponding pattern structure data to form a display deviation record, and returns the corresponding parameters for modification based on the deviation results.
[0021] S1. Obtain the pattern structure data of the garment to be modeled, divide the gradient levels according to the garment category, size specifications and pattern style, establish the pattern gradient parameter system, and construct the pattern gradient modeling task based on the pattern gradient parameter system and pattern structure data.
[0022] S1 first reads the contour control points, seam boundaries, and dimensional reference values of the garment to be modeled. Each dimensional reference value corresponds to a reference specification. For each target dimensional specification, the bust, waist, hip, garment length, and sleeve length are read, and the corresponding dimensional items in the reference specification are also read. The difference between the target dimensional specification and the reference specification on the same dimensional item is taken as the specification variation of that dimensional item.
[0023] Specification variations are allocated to contour control points according to the grading direction corresponding to the garment category. Bust differences are allocated to control points related to the bust contour, waist differences to control points related to the waist contour, and hip, garment length, and sleeve length are processed using the same principle. The allocation results from each control point are merged into the size difference for the current target size specification; subsequent processing only reads the size difference corresponding to the current gradient level. The pattern style processing reads the adjustments applicable to the current pattern style from the looseness allowance, dart allowance, and silhouette offset.
[0024] The ease allowance is obtained from the garment size and the wearer's net body measurements. First, read the garment's chest, waist, and hip measurements corresponding to the pattern structure data. Then, read the wearer's net chest, waist, and hip measurements corresponding to the current size specification. Subtract the corresponding wearer's net body measurement from the garment size for the same area, and the difference is used as the ease allowance for that area. Chest measurements are compared only with chest measurements, waist measurements are compared only with waist measurements, and hip measurements are compared only with hip measurements.
[0025] The dart amount is obtained from the dart endpoints. The dart endpoints in the pattern structure data are read, and the positions of the endpoints on both sides of the same dart on the corresponding seam panel boundary are determined. The distance between the endpoints on both sides is calculated along the seam panel boundary. The dart distances belonging to the same area are accumulated, and the accumulated result is used as the dart amount for that area.
[0026] The silhouette offset is obtained from the corresponding position of the current pattern silhouette and the baseline pattern silhouette. First, the baseline pattern silhouette with the same garment category and size specifications is read, and then the silhouette control point corresponding to the current pattern style is read at the same vertical baseline position. The displacement of the silhouette control point relative to the baseline pattern silhouette is determined along the normal direction of the baseline pattern silhouette, and the displacement direction is recorded. This normal displacement and direction are used as the silhouette offset.
[0027] After obtaining the allowance, dart allowance, and silhouette offset, determine the adjustment position and direction based on the correspondence between each adjustment item and the contour part. When multiple adjustment items exist for a contour part, first calculate the cumulative adjustment amount for that part, then check whether the boundaries of adjacent seam pieces after adjustment are still within the displacement range that allows for continuous splicing. The cumulative adjustment amount must not exceed this displacement range.
[0028] The size difference and shape adjustment amount are correlated with the gradient level to form a pattern gradient parameter system.
[0029] Please see Figure 2 S1 continues to execute the pattern gradient modeling task. Size differences are written to the corresponding contour control points, and shape adjustments are written to the corresponding contour control points and seam boundaries, obtaining the pattern structure data corresponding to the current gradient level. Subsequently, the garment category, gradient level, and this pattern structure data are associated as a task identifier. Each task identifier corresponds to a pattern gradient modeling task.
[0030] The task identifier remains unchanged in subsequent processing. The region modeling accuracy configuration generated by S2, the gradient template 3D model generated by S3, and the display deviation record generated by S4 all correspond to this task identifier, which can prevent the region accuracy data of one gradient level from being written to another gradient level.
[0031] S2 displays the pattern structure data corresponding to each pattern gradient modeling task in naked-eye 3D stereoscopic display, determines the structural complexity of the overall outline area, detail texture area, sewing structure area and fabric fold area, sorts them according to structural complexity, matches high, medium or low modeling resolution to each area, and determines the area modeling accuracy configuration.
[0032] S2 reads the pattern structure data corresponding to the current task identifier. The outline boundary, texture boundary, stitching line, and fold ridge line are converted into spatial line elements and mapped to the same naked-eye 3D display coordinate system.
[0033] Spatial line elements corresponding to contour boundaries are assigned to the overall contour area; spatial line elements corresponding to texture boundaries are assigned to the detail texture area; spatial line elements corresponding to sewing lines are assigned to the sewing structure area; and spatial line elements corresponding to fold ridges are assigned to the fabric fold area. After classification, the actual density of line elements, the number of direction changes, the number of overlaps, and the number of depth changes in each area are recorded to form a regional feature record.
[0034] The overall contour region is analyzed by reading the number of contour control points and the number of directional changes of adjacent contour segments. The number of contour control points is converted into control point density based on the contour length, and the number of directional changes is converted into directional change density based on the contour length. Control point density and directional change density serve as indicators of the complexity of the overall contour region.
[0035] The length of the texture boundary and the number of local depth abrupt changes in the detailed texture region are read, and then converted into feature density per unit area according to the area of the detailed texture region.
[0036] The sewing structure region reads the sewing thread path density and the number of path intersections, and then forms the corresponding unit area feature density according to the area of the sewing structure region.
[0037] The density of crease ridges and valleys and the number of local depth extremes in the fabric folded area are read, and then the corresponding unit area feature density is formed according to the area of the fabric folded area.
[0038] The above four types of areas use their own existing structural features, without writing texture features into the overall outline area or sewing line paths into the fabric fold area.
[0039] Please see Figure 3 After the complexity indices for each region are formed, they undergo uniform scaling within the current pattern gradient modeling task. For the same complexity index, the maximum value among the data being compared is read, and this maximum value is used as the conversion benchmark to convert the corresponding index to a uniform numerical range. The converted indices are then grouped by region to form the structural complexity of each region.
[0040] After structural complexity is determined, regions are arranged from highest to lowest numerical value. The sorting results are used to match high, medium, or low modeling resolutions.
[0041] The target mesh side length is determined separately from the actual spacing in the pattern structure data. The spacing between adjacent contour control points, the minimum spacing of texture boundaries, the spacing between adjacent seam lines, and the spacing between adjacent fold ridge lines are read. If a certain type of spacing does not exist in the current pattern structure data, that type of data is not included in the minimum value filter. The actual spacings are compared, and the minimum spacing is selected.
[0042] The reference side length for high-resolution targets is no greater than half the target side length of this minimum spacing. The target side length for medium-resolution targets is twice the reference side length for high-resolution targets. The target side length for low-resolution targets is four times the reference side length for high-resolution targets.
[0043] The resolution levels obtained by sorting the structural complexity are written into each region along with the corresponding target mesh side lengths to form the region modeling accuracy configuration. When S3 reads the region modeling accuracy configuration, it can directly obtain the target mesh side length corresponding to the current region without recalculating the resolution level.
[0044] S3, based on the regional modeling accuracy configuration, allocates the gradient modeling tasks of each pattern to multiple parallel threads, constructs and stitches together 3D meshes of different regions according to the corresponding modeling resolution, and generates gradient pattern 3D models.
[0045] S3 reads the entire unprocessed region for the current task. For each region, it reads two data points: the region area and the target mesh edge length from the region modeling precision configuration. Based on the region area and the target mesh edge length, it determines the expected number of mesh cells for that region.
[0046] The estimated number of grid cells is written to the pending records for this region. During thread allocation, the estimated number of grid cells in the allocated regions of each thread's task queue is read, and the sum of the estimated grid cell counts in the queue is calculated. This sum is used as the thread load value for the corresponding thread.
[0047] For an unallocated region, compare the current thread load values of all threads and write the region to the task queue of the thread with the lowest thread load value. After writing, the estimated number of grid cells for that region is included in the thread's new thread load value. Subsequent regions continue to be allocated according to the updated thread load values. After all pending regions have been allocated, the task-thread mapping relationship is obtained.
[0048] Please see Figure 4 Each thread processes the regions within the queue according to the task thread mapping relationship. When processing a region, the thread reads the region's boundary points, internal structure lines, and target mesh edge length. Boundary points define the region boundaries of the triangular mesh, internal structure lines define the locations of structures to be preserved within the region, and the target mesh edge length defines the mesh scale of the current region. Based on these three types of input, triangular meshing is performed to generate a 3D mesh for the region.
[0049] Once a 3D mesh for a region is completed, that region is removed from the pending state. The estimated number of mesh cells for regions not yet being processed in each thread's task queue is then recalculated, and the thread load value is updated.
[0050] After the update, the average load value of all parallel threads is calculated, and the deviation between the load value of each thread and the average load value is compared. If the deviation does not exceed a preset load deviation threshold, each thread continues to process the existing queue. The preset load deviation threshold is determined based on the number of threads, the allowable load fluctuation ratio, or the historical task execution time.
[0051] When the deviation between the load value of any thread and the average load value of all parallel threads exceeds a preset load deviation threshold, the current highest-load queue and lowest-load queue are read. Then, regions that have not yet started processing are filtered from the highest-load queue, and the region with the largest expected number of grid cells is selected and migrated to the lowest-load queue. After the migration is complete, the thread load values of the two queues are recalculated, and the task thread mapping relationship is updated.
[0052] Regions that have already begun gridding will not be migrated, and regions that have already completed gridding will not be written back to the processing queue. Therefore, task migration only applies to regions that have not yet begun processing.
[0053] After all 3D meshes for the entire region are generated, region stitching begins. For two spatially adjacent 3D mesh regions, the vertex sequence on their common boundary is first read. Vertices whose positions correspond directly are aligned. When the number of vertices on the common boundary differs due to the different side lengths of the target meshes on both sides, transitional mesh cells are added at positions where direct one-to-one correspondence is not possible.
[0054] After vertex alignment and transition mesh reconstruction are completed at the common boundary, the 3D meshes of the two regions are merged. The remaining regions are then processed according to their spatial adjacency. Once all regions of the current task are stitched together, a gradient-patterned 3D model is output and written back to the current task identifier.
[0055] S4 displays the 3D models of each gradient pattern in real time using naked-eye 3D. Based on the display results, it adjusts the pattern gradient parameter system and the regional modeling accuracy configuration, and then re-executes the corresponding pattern gradient modeling task.
[0056] S4 reads the gradient template 3D model corresponding to the current task identifier and obtains the real-time view frames of the model from multiple display viewpoints. At the same time, it reads the template structure data used to generate the model and projects it according to the corresponding display viewpoints.
[0057] After the pattern structure data is projected, contour baselines and stitching baselines are formed. The model contour and stitching boundary are read from the real-time view frame. The model contour is compared with the contour baseline under the same display viewpoint to obtain the contour position deviation; the model stitching boundary is compared with the stitching baseline under the same display viewpoint to obtain the stitching boundary position deviation.
[0058] Each display viewpoint generates its own positional deviation record, and the result of one display viewpoint is not directly written into another display viewpoint.
[0059] The local surface dispersion is obtained from the display depth of adjacent grid faces. The display depth of adjacent grid faces at the current viewpoint is read, and the depth jumps between adjacent grid faces are checked. The depth jumps correspond to the local surface dispersion state of the current region and are recorded as the local surface dispersion degree.
[0060] Contour position deviation, stitch boundary position deviation, and local surface dispersion are all written into the display deviation record for the current round.
[0061] Please see Figure 5 After the deviation record is generated, it enters one of two processing paths.
[0062] The first processing path compares the contour position deviation and the stitching boundary position deviation with a preset deviation threshold. If any position deviation exceeds the preset threshold, the corresponding parameter in the pattern gradient parameter system is used based on the task identifier and gradient level, and the pattern gradient parameter system for that gradient level is adjusted. The adjusted parameters are then used to regenerate the pattern structure data corresponding to that gradient level.
[0063] The second processing path compares the local surface dispersion with a preset dispersion threshold. When the local surface dispersion of a certain region exceeds the preset dispersion threshold, the region is located based on the region correspondence of that local location in the gradient pattern 3D model, and the modeling resolution of that region is increased, while the region modeling accuracy configuration is updated. The updated region is reconstructed by reading the new target mesh edge length when S3 is re-executed.
[0064] If the positional deviation does not exceed a preset deviation threshold, parameter adjustment for that gradient level will not be triggered. If the local surface dispersion does not exceed a preset dispersion threshold, resolution improvement for that region will not be triggered. These two types of judgments are handled separately to avoid directly changing the mesh resolution due to positional deviation, and also to avoid directly rewriting the size difference or shape adjustment amount due to local surface dispersion.
[0065] After any adjustment is triggered, the adjusted data is written back to the current task identifier, and the corresponding pattern gradient modeling task is re-executed. During re-execution, tasks whose pattern gradient parameter system has changed read the adjusted pattern structure data; regions whose region modeling precision configuration has changed read the updated target mesh side lengths. Parameters that have been replaced in the previous round are not used as input in the current round.
[0066] After each re-execution, the contour position deviation, stitch boundary position deviation, and local surface dispersion of the current round are retrieved again, and the corresponding values from the previous round are read for comparison. The data from the previous round is used to record the changes between consecutive rounds. Whether to stop is determined based on the comparison results of the current round's data with the corresponding thresholds.
[0067] If the current round still has contour position deviation or stitch boundary position deviation greater than the preset deviation threshold, continue to adjust the pattern gradient parameter system of the corresponding gradient level. If the current round still has local surface dispersion greater than the preset dispersion threshold, continue to increase the modeling resolution of the corresponding region.
[0068] When the positional deviations of all contours and all stitching boundaries are no greater than the preset deviation threshold, and the dispersion of all local surfaces is no greater than the preset dispersion threshold, the re-execution of the current pattern gradient modeling task is stopped. The gradient pattern 3D model of the current round is used as the output model of this task.
[0069] When multiple gradient levels exist in the same batch, each task performs the above judgment separately. Once a task reaches the stopping condition, only the re-execution of that task is stopped, while other tasks that have not yet reached the stopping condition continue to be processed.
[0070] In the above method, the pattern structure data is obtained from the garment pattern; the size difference is obtained from the size difference between the target size specification and the baseline specification; the shape adjustment amount is obtained from the looseness amount, the dart amount, and the silhouette offset amount; the area modeling accuracy configuration is obtained from the area structure complexity and the minimum spacing of the actual structure; the thread load value is obtained from the expected number of mesh cells in the area to be processed in the thread task queue; and the display feedback data is obtained from the real-time view frame of the gradient pattern 3D model and the projection result of the corresponding pattern structure data.
[0071] After the size difference is written to the contour control point, the pattern size state corresponding to the target size specification is changed. After the shape adjustment amount is written to the contour control point and the seam boundary, the contour state corresponding to the pattern style is changed. The structural complexity is used to determine the modeling resolution adopted for the area. The target mesh side length is used for the actual triangular mesh division. The expected number of mesh units is used for thread allocation and migration. The display deviation record returns the pattern gradient parameter system and the area modeling accuracy configuration respectively.
[0072] The specific steps described above illustrate the implementation process of this invention. When the size data name, region data name, or task field name changes equivalently, as long as the data source, processing object, processing order, and output purpose remain unchanged, the same data relationship can be followed. Without departing from the technical solution described in this invention, using equivalent data organization forms, equivalent execution methods, or equivalent field names for the same technical object does not alter the pattern gradient processing, region multi-resolution modeling, parallel task allocation, and display feedback relationships described in this invention.
Claims
1. A naked-eye 3D visualization method for gradient batch parallel multi-resolution 3D modeling of clothing patterns, characterized in that... Includes the following steps: Obtain the pattern structure data of the garment to be modeled, divide the gradient levels according to the garment category, size specifications and pattern style, establish a pattern gradient parameter system, and construct a pattern gradient modeling task based on the pattern gradient parameter system and the pattern structure data. The pattern structure data corresponding to each pattern gradient modeling task is displayed in naked-eye 3D stereoscopic display. The structural complexity of the overall outline area, detail texture area, sewing structure area and fabric fold area is determined. According to the structural complexity, high, medium or low modeling resolution is matched for each area to determine the area modeling accuracy configuration. Based on the region modeling accuracy configuration, each of the pattern gradient modeling tasks is assigned to multiple parallel threads. Three-dimensional meshes of different regions are constructed and stitched together according to the corresponding modeling resolution to generate gradient pattern three-dimensional models. The three-dimensional models of each gradient pattern are displayed in real-time using naked-eye 3D. Based on the display results, the gradient parameter system of the pattern and the modeling accuracy configuration of the region are adjusted, and the gradient modeling task of the pattern is re-executed.
2. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 1, characterized in that, The steps for establishing the pattern gradient parameter system and the pattern gradient modeling task include: Obtain the garment pattern, extract the outline control points, seam boundaries and size reference values, determine the size difference of the size specification with reference to the reference specification, determine the shape adjustment amount of the outline part according to the pattern style, establish the parameter mapping relationship between the gradient level and the size difference and the shape adjustment amount, and generate the pattern gradient parameter system; The size difference and the shape adjustment amount are written into the corresponding contour control points and seam boundaries to form the pattern structure data corresponding to each gradient level. The clothing category, the gradient level and the pattern structure data are associated as a task identifier to generate the pattern gradient modeling task.
3. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 2, characterized in that, The process for determining the size difference and the shape adjustment amount is as follows: Using the reference specification corresponding to the size reference value as a reference, read the chest, waist, hip, garment length and sleeve length of the target size specification, calculate the difference between the target size specification and the reference specification in each size item, and allocate the difference to the contour control point according to the grading direction of the garment category to generate the size difference; Obtain the applicable adjustment items from the looseness amount, dart amount, and silhouette offset amount corresponding to the pattern style, determine the adjustment position and direction based on the pre-stored correspondence between the adjustment items and the outline parts, and limit the cumulative adjustment amount of the outline parts within the displacement range of maintaining continuous splicing of adjacent seam boundaries to generate the shape adjustment amount; Specifically, the garment's chest, waist, and hip measurements corresponding to the pattern structure data are obtained, the human body net measurements corresponding to the size specifications are read, and the difference between the garment size and the human body net measurements for the same part is determined as the looseness allowance. The endpoints of the darts in the pattern structure data are analyzed, the spacing between the endpoints on both sides of each dart along the corresponding seam boundary is calculated, and the spacing is accumulated according to the corresponding area to generate the dart collection amount; Obtain a baseline pattern outline that is the same as the clothing category and the size specification. Calculate the normal displacement of the pattern style corresponding outline control point relative to the baseline pattern outline at the same vertical reference position, record the displacement direction, and generate the silhouette offset.
4. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 1, characterized in that, The steps for generating the region modeling accuracy configuration include: Obtain the pattern structure data corresponding to each pattern gradient modeling task, convert the contour boundary, texture boundary, sewing line and fold ridge line into spatial line elements and map them to the naked-eye 3D display coordinate system, and classify them into the overall contour area, the detailed texture area, the sewing structure area and the fabric fold area according to the structural attributes of the spatial line elements, and record the line element density, direction change times, overlap times and depth change times of each area to generate regional feature records; Based on the actual structural attributes of each region, a complexity index is selected from the region feature record. The complexity indices within the same gradient modeling task are converted to a unified numerical range according to their maximum values and added together to determine the structural complexity. Regions are arranged from high to low according to the structural complexity, and high modeling resolution, low modeling resolution and medium modeling resolution are matched to generate the region modeling accuracy configuration.
5. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 4, characterized in that, The specific complexity metrics for each region are as follows: The number of contour control points in the overall contour region and the number of times the direction of adjacent contour line segments changes are obtained. The control point density and the direction change density are determined according to the unit contour length, and the control point density and the direction change density are determined as the overall contour complexity index. The texture boundary length and number of local depth abrupt change positions of the detailed texture region, the sewing line path density and number of path intersection positions of the sewing structure region, and the fold ridge valley line density and number of local depth extreme value positions of the fabric fold region are obtained. They are converted into feature density per unit area according to the corresponding area, and the detailed texture complexity index, sewing structure complexity index, and fabric fold complexity index are generated.
6. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 4, characterized in that, The process of generating the region modeling accuracy configuration also includes: The spacing between adjacent contour control points, minimum spacing of texture boundaries, spacing between adjacent sewing lines, and spacing between adjacent fold ridges and valleys are obtained from the pattern structure data. The minimum spacing is selected from the actual existing spacing data, and the grid target side length that is not greater than half of the minimum spacing is determined as the high-resolution reference side length. The high-resolution baseline side length is determined as twice the medium-resolution target side length, and the high-resolution baseline side length is determined as four times the low-resolution target side length. The high-resolution baseline side length, the medium-resolution target side length, and the low-resolution target side length are written into the modeling accuracy parameters of the corresponding regions to generate the region modeling accuracy configuration.
7. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 1, characterized in that, The steps for generating the gradient pattern 3D model include: Obtain the region modeling accuracy configuration corresponding to each of the above-mentioned pattern gradient modeling tasks, determine the expected number of grid cells based on the region area and the target grid side length, allocate the region to be processed to the thread task queue with the lowest load according to the number of grid cells to be processed in each parallel thread, and generate a task thread mapping relationship. Each of the parallel threads reads the boundary points and internal structure lines of the corresponding region, determines the side length of the target mesh according to the corresponding modeling resolution, and performs triangular meshing to generate a three-dimensional mesh of the region. Obtain the sequence of common boundary vertices of the spatially adjacent 3D meshes of the region, align vertices at the same position and add transition mesh cells, merge the 3D meshes of the region that have completed boundary alignment, and generate the gradient pattern 3D model.
8. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 7, characterized in that, The process of generating the task thread mapping relationship is as follows: Read the expected number of grid cells in the allocated area of each thread task queue, determine the thread load value according to the sum of the expected number of grid cells in the queue, and write the area to be allocated into the thread task queue with the lowest thread load value; After each 3D mesh of the region is constructed, the thread load value of the corresponding parallel thread is updated. When the deviation of any thread load value from the average load value of all parallel threads exceeds a preset load deviation threshold, the region to be processed that has not yet started processing and has the largest expected number of mesh units in the highest load queue is moved to the lowest load queue, and the task thread mapping relationship is updated.
9. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 1, characterized in that, The specific process for processing the displayed results is as follows: The real-time view frames of each gradient pattern 3D model in multiple display viewpoints are obtained. The pattern structure data is projected according to the corresponding display viewpoint to generate contour baselines and stitching baselines. The positional deviation of the model contour and stitching boundary relative to the contour baselines and stitching baselines is calculated. The local surface dispersion is determined according to the display depth jump of adjacent mesh surfaces, and a display deviation record is generated. When the positional deviation exceeds a preset deviation threshold, the pattern gradient parameter system of the corresponding gradient level is adjusted. When the local surface dispersion exceeds a preset dispersion threshold, the modeling resolution of the corresponding region is increased and the region modeling accuracy configuration is updated. The corresponding pattern gradient modeling task is then re-executed.
10. The naked-eye 3D visualization method for clothing pattern gradient batch parallel multi-resolution 3D modeling according to claim 9, characterized in that, The specific stopping condition for re-executing the pattern gradient modeling task is as follows: After each re-execution of the pattern gradient modeling task, the contour position deviation, stitching boundary position deviation, and local surface dispersion in the current round of naked-eye 3D real-time display results are obtained and compared with the corresponding values in the previous round. When all the contour position deviations and the stitching boundary position deviations are not greater than a preset deviation threshold, and all the local surface dispersions are not greater than a preset dispersion threshold, the re-execution of the pattern gradient modeling task is stopped.
